Optimised row fill routines

[stevesims]

Currently the "set" paint mode "row fill" operations are fairly optimal, but the routines for the other paint modes are not.

With the exception of 8 colour modes (which we don't currently use in agon-vdp anyway) it should be fairly straightforward to allow the various operations to make use of 8-bit and 32-bit operations, rather than the current implementation that just does things a single pixel at a time.